Codex Luna Reserve (gpt-reserve): What It Is, Quotas & Troubleshooting
Luna Reserve (surfaced in Codex CLI telemetry and status outputs as gpt-reserve) is an automated fallback capacity mode provided by OpenAI for selected personal ChatGPT Plus and Pro accounts in supported Codex and ChatGPT Work environments. When your primary 5-hour rolling or weekly subscription allowance is exhausted, Luna Reserve grants additional, separate usage powered strictly by GPT-5.6 Luna to keep your coding session active. It is an included feature—not a paid add-on, separate purchase, or API credit pool—and it does not restore your normal allowance, unlock frontier models (like Sol or Astra), or apply to ChatGPT Business or Enterprise workspaces.
Strictly powers GPT-5.6 Luna (lightweight, high-throughput). It cannot execute tasks on GPT-5.6 Sol, Terra, or GPT-6 Astra.
Available to selected personal accounts during rollout. Excluded from ChatGPT Business and Enterprise workspaces.
Has its own capped usage limit and reset timing. Does not restore or modify your plan's primary weekly or 5-hour limits.
Included automatically as fallback capacity. You cannot purchase additional reserve capacity or convert it into API credits.
Surfaces in /status telemetry when provisioned, tracking remaining reserve percentage.
Displays a moon icon or banner in desktop and web warning dialogues when sessions switch to fallback reserve execution.
Why Did “gpt-reserve” Appear in /status?
Developers using the Codex CLI frequently discover Luna Reserve through terminal telemetry. When OpenAI provisions an account for fallback capacity, running /status reveals an extra allocation labeled gpt-reserve alongside your standard 5-hour rolling limit and weekly account percentage.
In desktop app and web versions of ChatGPT Work, this same state is indicated by a warning banner containing a moon icon, informing you that your primary model limits have been reached and that sessions are running on Luna Reserve.
Regular Codex Allowance vs. Luna Reserve
How fallback reserve capacity compares to regular quotas, banked promotional resets, and paid recovery options.
| Dimension | Regular Codex Quota | Luna Reserve (gpt-reserve) | Banked Reset | Paid Reset / Credits |
|---|---|---|---|---|
| Model Access | All plan models (GPT-6 Astra, GPT-5.6 Sol, Terra, Luna) | Strictly GPT-5.6 Luna (cannot run Sol, Terra, or Astra) | All plan models (restores full plan access) | All plan models (full plan access) |
| Trigger Condition | Active continuously while 5-hour and weekly quota remain > 0% | Engages as a fallback after primary regular limits are exhausted | Manual redemption in Settings > Usage when quota is low or exhausted | Instant checkout or automatic deduction when quota runs out |
| Allowance Pool | Shared subscription quota (5-hour rolling + weekly window) | Separate, isolated reserve allowance (shows as gpt-reserve) | Refreshes existing 5-hour and weekly windows to 100% | Accelerated weekly quota or metered prepaid token balance |
| Cost / Pricing | Included in subscription ($20/mo Plus, $100–$200/mo Pro) | Free ($0); included fallback benefit for eligible accounts | Free promotional grant ($0; earned via campaigns or referrals) | One-time checkout fee (paid reset) or metered credit purchase |
| Plan Eligibility | All active subscriptions (Free, Plus, Pro, Business, Enterprise) | Selected personal Plus and Pro accounts only (A/B rollout) | Qualifying promotional Plus and Pro accounts | Eligible personal Plus/Pro (paid reset) or Plus/Pro/Business (credits) |
| Workspace Support | Personal, Business Standard/Premium, and Enterprise | Personal accounts only (Business & Enterprise strictly excluded) | Personal accounts only (distinct coworker invites for Business) | Personal (paid reset) or Workspace Billing (shared team credits) |
| Reset Impact | Normal 5-hour rolling recovery + scheduled weekly reset anchor | Has its own usage limit; returns to regular models once normal quota resets | Instantly sets quota to 100% and establishes a new weekly reset date | Paid reset establishes a new weekly cycle; credits draw down as consumed |
Eligibility & Exclusions: Who Gets Luna Reserve?
OpenAI maintains strict separation between personal subscriptions and organization workspaces. Luna Reserve is explicitly restricted to personal accounts and is subject to dynamic availability.
Available to select personal Plus accounts on supported Codex and ChatGPT Work desktop and web interfaces. Subject to rollout availability.
Available to select personal Pro accounts. Provides fallback GPT-5.6 Luna capacity once primary high-tier quotas are exhausted.
Explicitly excluded by OpenAI documentation. Business workspaces manage capacity through seat allowances, pooling, or shared workspace credits.
Explicitly excluded by OpenAI documentation. Enterprise accounts use contract-governed custom pools and administrative spend limits.
Luna Reserve is an included feature for select paid personal subscriptions, not available to free tier accounts.
API usage is billed per token and governed by organization TPM/RPM rate limits, completely separate from subscription reserve pools.
OpenAI Help Center documentation explicitly states: “Availability can change and depends on your account status and supported app version. Reaching a regular usage limit does not automatically guarantee access to the reserve feature.” Access may vary by region, account age, and active testing cohorts.
Reset Mechanics: What Happens When Regular Quota Returns?
Because regular usage and Luna Reserve have independent allowances, exhausting one does not change the recovery timeline of the other:
1While on Luna Reserve
Your regular models (Sol, Terra, Astra) remain locked. Your background 5-hour rolling recovery and weekly anchor timer continue counting down on their normal schedules. Using Luna Reserve does not delay or pause your regular reset clock.
2When Regular Limits Refresh
As soon as your 5-hour window recovers or weekly reset completes, full model access is restored. You can immediately return to using GPT-5.6 Sol or GPT-6 Astra without penalty. The reserve meter goes dormant until regular limits are hit again.
Troubleshooting: Reserve Meter Shows Usage but Codex Won't Run
First confirm that Luna Reserve is active for your current account and that you are using a supported surface (desktop app or ChatGPT Work on the web). Model-selection and client-state issues have been reported; switching models or restarting may help, but is not guaranteed.
Observed cause: Luna Reserve strictly executes on gpt-5.6-luna. If your current project, command-line arguments, or subagent configuration (such as in .codex/agents/*.toml) specifies gpt-5.6-sol or gpt-6-astra, Codex will pause and return a usage limit error because your allowance for those models is 0%.
Run /model gpt-5.6-luna in the CLI prompt, or select Luna from your model picker. Switching to Luna may help route requests to your remaining gpt-reserve quota, but is not guaranteed.
Observed cause: Luna Reserve is not an infinite safety buffer. Like regular quotas, it has a finite ceiling. If heavy agentic workflows consume the entire reserve pool (gpt-reserve: 0%), you may need to wait for regular usage to return or use another option available on your account, such as credits or a reset.
Run /status to verify gpt-reserve percentage. If 0%, your options are: (1) wait for regular usage to return (rolling 5h or weekly recovery), (2) apply a banked reset in Settings > Usage, (3) buy a paid instant reset, or (4) switch to metered API billing via OPENAI_API_KEY.
Observed community report: Multiple developers on GitHub and community forums have reported cases where the client remains locked in a quota-exhausted state immediately after primary limits hit zero, failing to auto-negotiate fallback routing.
Quit and restart the Codex desktop application, or exit and relaunch your Codex CLI session. Restarting may help the client re-authenticate with OpenAI telemetry endpoints and pick up the active reserve routing profile, but is not guaranteed.
Documented behavior: OpenAI documentation clarifies that if you reach a usage limit mid-turn, Codex can continue working on that active turn subject to fair-use limits. Once that turn concludes, subsequent prompts pause until routed through Luna Reserve or refreshed.
If switching to Luna and restarting do not resolve the issue, capture your usage indicator (or /status output), app version, and timestamp, and contact OpenAI Support.
How This Page Is Sourced
Product rules, plan exclusions (such as Business and Enterprise workspace restrictions), model limitations (strict GPT-5.6 Luna constraint), and separate allowance policies are sourced directly from official OpenAI documentation in the Help Center article “Luna Reserve in Codex and ChatGPT Work” (help.openai.com). CLI telemetry patterns (such as gpt-reserve) and troubleshooting workarounds reflect verified developer test sessions and qualified user reports from the Codex developer community.
Frequently Asked Questions
Direct answers to the most common questions about Codex Luna Reserve and gpt-reserve telemetry.
What is Luna Reserve in Codex?
Luna Reserve is an automated fallback feature in OpenAI Codex and ChatGPT Work. When an eligible personal account exhausts its regular 5-hour rolling or weekly subscription quota, Luna Reserve provides additional usage specifically powered by GPT-5.6 Luna. It allows ongoing coding tasks and conversations to continue rather than pausing execution immediately.
Why did gpt-reserve appear in my Codex /status output?
In the Codex CLI, OpenAI surfaces backend fallback allocations under the metric name 'gpt-reserve'. When your account is provisioned with Luna Reserve, /status displays your regular allowance meters alongside this dedicated reserve pool. When your primary allowance hits 0%, the client attempts to route tasks to this reserve tier.
Is Luna Reserve free or does it cost extra?
Luna Reserve is free ($0). It is an included platform fallback capability for eligible accounts, not an add-on, separate purchase, or API credit pool. OpenAI does not allow users to purchase additional Luna Reserve capacity.
Who is eligible for Luna Reserve?
According to official OpenAI Help Center documentation, Luna Reserve is available only to select personal ChatGPT Plus and Pro accounts using supported Codex and ChatGPT Work experiences (desktop app or ChatGPT Work on the web). It is explicitly not available in ChatGPT Business or Enterprise workspaces.
Does Luna Reserve have a separate usage limit?
Yes. OpenAI documentation specifies that regular usage and Luna Reserve operate on separate allowances. Luna Reserve is not unlimited; it has its own distinct usage limit. If you exhaust your reserve allowance, you may need to wait for regular usage to return or use another option available on your account, such as credits or a reset.
Does Luna Reserve restore my normal Codex allowance?
No. Luna Reserve does not restore your regular 5-hour or weekly allowance, nor does it unlock your plan's higher-tier models such as GPT-5.6 Sol or GPT-6 Astra. It serves strictly as a safety net to run GPT-5.6 Luna until your normal allowance refreshes.
What happens when my regular Codex usage resets?
Once your scheduled weekly reset arrives or your rolling 5-hour window recovers, your regular allowance becomes available again. At that point, you return to the standard models and quota limits included with your subscription plan.
Why does the reserve meter show usage, but Codex will not run?
First confirm that Luna Reserve is active for your current account and supported surface (desktop app or ChatGPT Work on the web). Model-selection and client-state issues have been reported—such as sessions remaining pinned to GPT-5.6 Sol or GPT-6 Astra, which Luna Reserve cannot execute, or stale client session caches. Switching to GPT-5.6 Luna (e.g. via '/model gpt-5.6-luna') or restarting the client application may help, but is not guaranteed. If the issue persists, capture your usage indicator, app version, and timestamp and contact OpenAI Support.
Why does Luna Reserve not appear on my account?
There are three common reasons: (1) Workspace type: You are logged into a ChatGPT Business, Enterprise, or Team workspace, which OpenAI explicitly excludes. (2) Phased rollout: OpenAI notes that access is selective, subject to change, and not guaranteed for every account upon hitting a limit. (3) Surface: You are using standard consumer ChatGPT web chat rather than supported Codex or ChatGPT Work surfaces.
Can I use Luna Reserve with custom agent workflows?
Yes, provided the agent is configured to use GPT-5.6 Luna. If an agent configuration (such as in .codex/agents/*.toml) specifies GPT-5.6 Sol, GPT-6 Astra, or another non-Luna model, the agent will pause with an exhausted quota error once your regular allowance runs out.
Review all base plan quotas and model allocations across Plus, Pro, and Business in the Codex usage limits table.
Wondering when your primary 5-hour rolling or weekly limits refresh? Read when does Codex usage reset.
Have a promotional grant stored in your account? Read our complete guide to Codex banked resets.
Completely tapped out of regular quota and Luna Reserve? Compare Codex credits, paid instant resets, and API billing to get back to coding immediately.
Using GPT-6 Astra? Read how frontier reasoning draws from your shared allowance in our GPT-6 Astra Codex limits guide.
Stuck on error codes or throttled sessions? Review our Codex rate limit and 429 triage guide.